Overlapped image target recognition and positioning algorithm based on sub-field shape modulation coding

Abstract

The present application relates to a kind of based on the shape modulation coding of sub-field of view aliasing image target recognition and positioning algorithm, it is related to aliasing image target recognition, field of view determination and positioning technical field, solve the technical problem that the scale of detector and field of view cannot be considered under the condition of limited detector in prior art when realizing large field of view, high-resolution imaging;The algorithm includes the following steps: foreground target recognition;Target rotation state tracking;Target motion trajectory and spot shape change trend correlation discrimination and actual spatial position inverse solution.The present application is based on the shape modulation coding of sub-field of view aliasing image target recognition and positioning algorithm, with the advantage of small target surface detector low power consumption, low cost, realize space-based early warning large width detection, break through the bottleneck that the large field of view imaging of infrared optical system is limited to the manufacturing process of infrared detector.The present application is realized by the extraction of continuous frame spot shape feature and trajectory-shape dual-domain correlation modeling, the unique determination of the field of view of target space and high-precision positioning.

Description

Technical Field

[0001] This invention relates to the fields of target recognition, field of view determination and localization in aliased images, and particularly to an algorithm for target recognition and localization in aliased images based on field-of-view shape modulation coding. Background Technology

[0002] Space-based infrared early warning systems are a crucial application scenario for space optical detection, widely used for on-orbit monitoring of high-temperature radiation sources on the ground and in the air, especially the typical infrared signature signals generated during missile launches. In recent years, with continuous advancements in satellite platforms, launch vehicles, and infrared detector manufacturing processes, constructing space-based infrared detection systems with high resolution and wide coverage has become possible. However, limited by the current detector array size, pixel scale, and detection sensitivity, the system still faces bottlenecks in expanding its field of view (FOV). Limited detector arrays often cannot simultaneously achieve high-resolution imaging and wide-area coverage, making it difficult for a single satellite to continuously monitor large areas. Therefore, how to achieve large field of view and high-resolution imaging under limited detector conditions has become a pressing technical challenge. The principle of field-of-view shape modulation coding-detector multiplexing is as follows... Figure 1 As shown, the optical system structure is as follows Figure 2 As shown.

[0003] Existing research has attempted to overcome the contradiction between detector scale and field of view by proposing various alternative solutions. For example, pushbroom imaging relies on satellite orbital motion to achieve wide-area coverage, but this method places extremely high demands on platform attitude stability and on-orbit data processing capabilities, and targets are prone to missed detection due to insufficient time sampling during the imaging cycle. While multi-satellite networking can alleviate the inadequacy of single-satellite coverage to some extent, the implementation costs of its constellation size, on-orbit formation control, and high-speed data transmission links are considerable. Although the development of large-area high-sensitivity detectors can directly expand the system's field of view, it is still difficult to meet the needs of engineering applications due to process difficulty and manufacturing costs. In addition, some methods attempt to achieve wide-area monitoring through optical structure optimization or multi-focal plane detection, but in complex target scenarios, it is often difficult to balance resolution, sensitivity, and real-time performance. Although these methods have their own achievements, they generally suffer from high implementation complexity, poor engineering adaptability, or excessive costs, which are insufficient to support large-scale applications and cannot fundamentally overcome the bottlenecks in detector manufacturing processes. Summary of the Invention

[0004] This invention aims to solve the technical problem that the detector size and field of view cannot be simultaneously achieved when realizing large field of view and high-resolution imaging under limited detector conditions in the prior art, and provides an aliased image target recognition and localization algorithm based on field-of-view shape modulation coding.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] An algorithm for target recognition and localization in aliased images based on field-of-view shape modulation coding includes the following steps:

[0007] Step 1: Foreground target identification;

[0008] Identify foreground targets;

[0009] Step 2: Target rotation state tracking;

[0010] By fusing principal axis angle observation, angular velocity observation and second-order Kalman filtering, continuous estimation of target rotation is achieved;

[0011] Step 3: Determine the correlation between the target's motion trajectory and the trend of light spot shape change, and inversely solve the actual spatial location;

[0012] Key feature information of the target is extracted from each frame of the image to further determine the target's motion trajectory and light spot morphology changes. Through dual-domain correlation modeling of target motion trajectory and light spot morphology evolution, the unique determination and high-precision positioning of the target's object space field of view are achieved.

[0013] In the above technical solution, step 3, extracting the key feature information of the target includes: the position of the spot centroid, the pixel area it occupies, and the target's position within the target area. Dimensions in the direction.

[0014] In the above technical solution, step 2 specifically includes:

[0015] 2a) Observation of principal axis angle;

[0016] For the target region detected in the foreground mask, the set of pixel coordinates within the region is recorded and centered; a covariance matrix is ​​constructed.

[0017] Through eigenvalue decomposition The largest eigenvalue of the covariance matrix Corresponding feature vector The principal axis direction; principal axis angle Represented as:

[0018] ;

[0019] in, is an eigenvalue of the covariance matrix; Eigenvalues Corresponding feature vectors; for eigenvectors of direction; for eigenvectors of direction;

[0020] 2b) Angular velocity observation;

[0021] Select several stable feature points within the target area. Tracking is performed; feature points are calculated relative to the target centroid in each frame. Angle :

[0022] ;

[0023] in, Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target region coordinate; Select several stable feature points within the target region coordinate; and Representing the target centroid respectively coordinates and coordinate;

[0024] This gives us the instantaneous angular velocity of the target. :

[0025] , ;

[0026] in, For frame interval; For the first -1 frame feature points relative to the target centroid Angle; For the first Frame feature points relative to the target centroid Angle; Indicates the first One feature point, Indicates the total number of feature points;

[0027] 2c) Second-order Kalman filter fusion;

[0028] Will Provided principal axis angle observation and Using the provided angular velocity observations as input, a second-order Kalman filter is constructed to achieve joint estimation of the target rotation.

[0029] In the above technical solution, step 3, the dual-domain correlation modeling of target motion trajectory and light spot morphology evolution, specifically includes:

[0030] 3a) Modeling the motion of the center of mass;

[0031] Let the first The centroid of the target in the frame is Then its inter-frame displacement for:

[0032] ;

[0033] in, For the first Frame target centroid; Inter-frame displacement Directional components; Inter-frame displacement Directional components; These correspond to the horizontal and vertical movement trends, respectively.

[0034] 3b) Modeling of light spot morphology features;

[0035] Define the goal in the first Frame morphological parameter vector for:

[0036] ;

[0037] in: This represents the pixel area of ​​the target within the foreground mask; Indicates the light spot is in The scale of direction; This represents the principal axis rotation angle obtained by the target rotation state tracking algorithm;

[0038] Inter-frame difference characterizes the evolution trend of light spots:

[0039] ;

[0040] in, The change in the morphological parameter vector; For the first -1 frame's morphological parameter vector; The change in pixel area of ​​the target in the foreground mask; For the light spot in The magnitude of change in direction; The change in the spindle rotation angle; Indicates clockwise rotation; Indicates counterclockwise rotation;

[0041] 3c) Joint modeling of target motion trajectory and light spot morphology;

[0042] To correlate the target's motion trajectory with the evolution of the light spot morphology, a joint feature vector is defined. :

[0043] ;

[0044] The target is in the timing window. The overall evolution within is represented as:

[0045] ;

[0046] in, For the target in the timing window The overall evolution within; For a certain moment The evolution of the target; Timing window At a certain moment within;

[0047] 3d) Field of view region discrimination and inverse spatial location solution;

[0048] Suppose the field of view is divided into several non-overlapping regions. Then there exists a deterministic mapping relationship. :

[0049] ;

[0050] By identification The coupled characteristics of the target's motion trajectory and the light spot shape contained within it can directly determine the unique field of view to which the target belongs. ;

[0051] Based on the geometric correspondence between field of view partitions and object space :

[0052] ;

[0053] The target is derived from the observation results in the image plane to its actual spatial coordinates. .

[0054] In the above technical solution, step 1 specifically involves: using a Gaussian mixture model to model the background, modeling the historical value of each pixel as a combination of multiple weighted Gaussian distributions, with each distribution corresponding to the background state of the pixel at different times; if the current pixel value deviates significantly from these background distributions, it is determined to be a foreground target.

[0055] In the above technical solution, step 1 specifically involves using frame difference method or optical flow method to achieve foreground target recognition.

[0056] In the above technical solution, step 3: using The module extracts key feature information of the target in each frame of the image.

[0057] The present invention has the following beneficial effects:

[0058] The aliased image target recognition and localization algorithm based on field-of-view shape modulation coding of the present invention has an optical system with a coding modulation module at the front end. The coding modulation module is used to fold and compress the multi-field light rays that originally needed to cover a large-area array focal plane onto a single small-area array detector. The present invention achieves wide-swath detection for space-based early warning by leveraging the advantages of low power consumption and low cost of small target detectors, and breaks through the bottleneck of large-field imaging of infrared optical systems being limited by the manufacturing process of infrared detectors.

[0059] The aliased image target recognition and localization algorithm based on field-of-view shape modulation coding of the present invention is a spatiotemporally joint intelligent small target recognition and field-of-view discrimination localization algorithm. By extracting the morphological features of light spots in continuous frames and modeling the trajectory-shape dual-domain association, it realizes the unique determination and high-precision localization of the target's object space field of view.

[0060] The aliased image target recognition and localization algorithm based on field-of-view shape modulation coding of the present invention has the advantages of low power consumption and low cost of small target detector and large field of view coverage, breaking through the limitations of existing infrared focal plane manufacturing processes on field of view expansion. Attached Figure Description

[0061] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0062] Figure 1 This is a schematic diagram of the field-of-view shape modulation coding-detector multiplexing principle. The upper part shows how the cylindrical mirror modulates the field of view in sections using shape modulation, with the spot shape representing the modulation effect of different field points within each section presented in PSF format. The lower part illustrates the mechanism by which the detector multiplexing component folds the optical path of the system to enable a small-area detector to replace a large-area detector in receiving large-field-of-view imaging.

[0063] Figure 2 This is a schematic diagram of a traditional infrared system large field-of-view imaging technique in the prior art.

[0064] Figure 3This is a schematic diagram of the target recognition and localization algorithm for aliased images based on field-of-view shape modulation coding of the present invention. Wherein: a represents the target motion trajectory on the aliased image received by the detector after multiplexing; b represents the actual motion trajectory of the object in the space corresponding to different field regions; c represents the trend of the target's light spot change after modulation coding by the cylindrical mirror under this trajectory; d represents the actual trend of the target's light spot change received by the detector.

[0065] Figure 4 This diagram illustrates the simulation and result analysis of the target rotation state tracking algorithm. Specifically: a) Image sequences at different times in the target motion video received by the detector; b) Gaussian and salt-and-pepper noise added; c) Gaussian and median filtering for noise reduction; d) Foreground target recognition result; e) Spot rotation state tracking of the identified foreground target; f) Field of view partitioning based on the coupling relationship between the spot rotation state and the actual target trajectory in the tracked image sequence (the image is for example, showing only five image sequences); g) System detection rate at different frame intervals (5-15 frames). ) and false alarm rate ( Statistical results of changes with frame interval and receiver operating characteristics plotted based on this data ( )curve.

[0066] Figure 5 This diagram illustrates the correlation between target motion trajectory and spot shape change trends in different field-of-view zones, and the positioning results. In this diagram: a) demonstrates the application of the algorithm of this invention to targets in different field-of-view regions; b) represents the inverse solution of the target's actual spatial position, from top to bottom representing the field-of-view zone, the positioning data from the inverse solution of the object's spatial position, and the actual spatial position of the target.

[0067] Figure 6 This is a schematic diagram showing the results of the field-of-view positioning accuracy error analysis. Where: a is... Histogram of orientation angle error distribution and Gaussian fitting results; b is The cumulative distribution function (CDF) of orientation error; c is The curve showing the variation of root mean square error (RMS) in orientation with frame number; d is... Histogram of orientation angle error distribution and Gaussian fitting results; e is The cumulative distribution function (CDF) of the orientation error; f is Curve showing the change of root mean square error (RMS) in direction with frame number. Detailed Implementation

[0068] The aliased image target recognition and localization algorithm based on field-of-view shape modulation coding of the present invention has an optical system with a coding modulation module at the front end. The coding modulation module is used to fold and compress the multi-field light rays that originally needed to cover a large-area array focal plane onto a single small-area array detector, thereby significantly reducing the detector size and manufacturing difficulty.

[0069] Furthermore, this invention constructs a spatiotemporal joint intelligent small target recognition and field-of-view discrimination localization algorithm. By extracting the morphological features of light spots in continuous frames and modeling the trajectory-shape dual-domain association, it achieves unique determination and high-precision positioning of the target's object space field of view.

[0070] The solution of this invention combines the advantages of low power consumption and low cost of small target detectors with the ability to cover a large field of view. It breaks through the limitations of existing infrared focal plane array manufacturing processes on field of view expansion and provides a new technical path with engineering feasibility and scalability for space-based infrared early warning systems.

[0071] The present invention will now be described in detail with reference to the accompanying drawings.

[0072] like Figure 3 As shown, the target recognition and localization algorithm for aliased images based on field-of-view shape modulation coding of the present invention is implemented in MATLAB to identify deformable small targets in multiplexed aliased images after field-of-view partition coding under detector multiplexing. The specific steps are as follows:

[0073] Step 1: Foreground target identification;

[0074] A Gaussian Mixture Model (GMM) is used for background modeling to identify small moving targets in the foreground. The historical values ​​of each pixel are modeled as a combination of multiple weighted Gaussian distributions, each corresponding to the background state of the pixel at different times. If the current pixel value deviates significantly from these background distributions, it can be identified as a foreground target.

[0075] In other specific implementations, the foreground target recognition method may also employ frame difference method or optical flow method.

[0076] Step 2: Target rotation state tracking;

[0077] After foreground target identification is completed, in order to achieve accurate tracking of the rotation angle of small-scale foreground targets, the method of this invention designs a multi-stage fusion angle estimation method. Specifically, after foreground segmentation is completed, the target rotation is continuously estimated by fusing principal axis angle observation, angular velocity observation, and second-order Kalman filtering.

[0078] like Figure 4 As shown, the specific steps are as follows:

[0079] 2a) Principal axis angle observation ( Ellipse fitting);

[0080] For the target region detected in the foreground mask, the set of pixel coordinates within the region is recorded and centered; a covariance matrix is ​​constructed.

[0081] Through eigenvalue decomposition The largest eigenvalue of the covariance matrix Corresponding feature vector This refers to the principal axis direction of the target. Principal axis angle. It can be represented as:

[0082]

[0083] in, is an eigenvalue of the covariance matrix; Eigenvalues Corresponding feature vectors; for eigenvectors of direction; for eigenvectors of direction;

[0084] This principal axis angle As a static rotation observation of the target in each frame, it provides a reference for rotation trend analysis.

[0085] 2b) Angular velocity observation ( Optical flow tracing);

[0086] To obtain the rotational change information of the target between consecutive frames, a method based on... Optical flow ( Angular velocity observation.

[0087] Select several stable feature points within the target area. Tracking is performed. Feature points are calculated relative to the target centroid in each frame. Angle :

[0088]

[0089] in, Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target region coordinate; Select several stable feature points within the target region coordinate; and Representing the target centroid respectively coordinates and coordinate;

[0090] This gives us the instantaneous angular velocity of the target. (Unit is) ):

[0091] ,

[0092] in, For frame interval; For the first -1 frame feature points relative to the target centroid Angle; For the first Frame feature points relative to the target centroid Angle; Indicates the first One feature point, Indicates the total number of feature points;

[0093] This angular velocity observation can capture the minute rotations of small targets, providing supplementary information for dynamic angle estimation.

[0094] 2c) Second-order Kalman filter fusion (principal axis angle + angular velocity);

[0095] Will Provided principal axis angle observation and Using the provided angular velocity observations as input, a second-order Kalman filter is constructed to achieve joint estimation.

[0096] By using the standard Kalman prediction-update formula, the target angle is smoothly estimated in the presence of noise or illumination changes on small-scale targets, while tracking its rotation speed to achieve continuous rotation tracking.

[0097] Step 3: Determine the correlation between the target's motion trajectory and the trend of light spot morphology changes, and inversely solve the actual spatial location;

[0098] Use in each frame image The module extracts key feature information of the target, including the position of the spot centroid, the pixel area it occupies, and the target's position within the target area. The dimensions in the direction are used to further determine the target's motion trajectory (i.e., the target's dynamic behavior) and the changes in the light spot morphology. By modeling the dual-domain correlation between the target's motion trajectory and the light spot morphology evolution, the unique determination and high-precision positioning of the target's object space field of view can be achieved.

[0099] The target's trajectory is described by the inter-frame displacement of its center of mass in the horizontal and vertical directions, such as moving left or right, or drifting up or down; while the evolution of the light spot's shape is described by the rotation angle (clockwise / counterclockwise) along the principal axis. It is depicted by dimensional changes (increase / decrease) in the direction and changes in pixel area.

[0100] like Figure 5 As shown, the mathematical modeling of the dual-domain correlation between the target's motion trajectory and the evolution of the light spot morphology is as follows:

[0101] 3a) Modeling the motion of the center of mass;

[0102] Let the first The centroid of the target in the frame is Then its inter-frame displacement for:

[0103]

[0104] in, For the first Frame target centroid; Inter-frame displacement Directional components; Inter-frame displacement Directional components; These correspond to the movement trends in the horizontal and vertical directions, respectively.

[0105] 3b) Modeling of light spot morphology features;

[0106] Define the goal in the first Frame morphological parameter vector for:

[0107]

[0108] in: This represents the pixel area of ​​the target within the foreground mask; Indicates the light spot is in The scale of direction; This represents the spindle rotation angle obtained by the target rotation state tracking algorithm.

[0109] Inter-frame difference characterizes the evolution trend of light spots:

[0110]

[0111] in, The change in the morphological parameter vector; For the first -1 frame's morphological parameter vector; The change in pixel area of ​​the target in the foreground mask; For the light spot in The magnitude of change in direction; The change in the spindle rotation angle; Indicates clockwise rotation; It indicates counterclockwise rotation.

[0112] 3c) Joint modeling of target motion trajectory and light spot morphology evolution;

[0113] To correlate the target's motion trajectory with the evolution of the light spot morphology, a joint feature vector is defined. :

[0114]

[0115] The target is in the timing window. The overall evolution within can be represented as:

[0116]

[0117] in, For the target in the timing window The overall evolution within; For a certain moment The evolution of the target; Timing window At a certain moment within.

[0118] This joint feature set describes the coupling relationship between the target's motion trajectory and the light spot morphology.

[0119] 3d) Field of view region discrimination and inverse spatial location solution;

[0120] The joint feature set can simultaneously reflect the changing trends of the target's motion trajectory and the spot morphology. Because targets within different fields of view exhibit unique and stable patterns in terms of motion direction, spot rotation, and scale changes, therefore... A one-to-one correspondence is established between the target and the spatial region to which it belongs.

[0121] Specifically, suppose the field of view is divided into several non-overlapping regions. Then there exists a deterministic mapping relationship. :

[0122]

[0123] That is, through identification The coupled characteristics of the target motion trajectory and light spot shape contained within the image can directly determine the unique field of view to which the target belongs. .

[0124] Furthermore, based on the geometric correspondence between the field of view partitions and the object space... :

[0125]

[0126] This allows the target to be derived from the observation results in the image plane and converted to its actual spatial coordinates. Therefore, based on the joint discrimination mechanism of temporal trajectory features and light spot morphology features, the unique location of the target in the object space is achieved.

[0127] In other specific implementations, the target centroid extraction method can be modified when resolving the actual spatial location of the target.

[0128] like Figure 4 and 6 As shown in the simulation results, the system achieved a detection rate of 94.04% ( ), 5.96% false negative rate ( ) and a false alarm rate of 1.79% ), The mean of the orientation angle positioning error is -1.39 arcsec, the standard deviation is 7.80 arcsec, and the RMS is 7.91 arcsec. The mean of the directional angle positioning error is 13.32 arcsec, the standard deviation is 15.11 arcsec, and the RMS is 20.14 arcsec. Both directional errors exhibit an approximately Gaussian distribution, and the histogram and fitted curve show good agreement. The CDF curve indicates that over 90% of the sample angle errors fall within ±2 arcsec. Within the specified range, the positioning results demonstrate high stability and consistency. The RMS changes smoothly with the frame sequence, without significant abrupt changes, indicating that the algorithm of this invention maintains strong robustness throughout the entire frame sequence and is not easily affected by individual abnormal frames. Based on the geometric relationship between the LEO orbit and the imaging system, the angle error is converted into ground resolution, yielding... The mean error of the directional ground positioning is 3.4m, and the RMS is 19.3m; The mean ground positioning error was 32.3m, and the RMS was 48.8m. This result demonstrates that the algorithm of this invention can achieve angular positioning accuracy from sub-arcsecond to tens of arcseconds, and also possesses quantifiable meter-level positioning capability on ground projection, meeting the engineering requirements of space-based infrared early warning systems for initial positioning of high-temperature targets on the ground and in the air. Experimental results show that under real observation conditions, the algorithm of this invention can still stably identify moving targets and accurately determine their field-of-view attribution. In a total of 10,800 frames of image sequences in the experimental video, the algorithm of this invention achieved a detection rate of 93.86% (…). ) and a false negative rate of 6.14% ) and a false alarm rate of 1.83% ( The results are highly consistent with the simulation results (94.04% / 5.96% / 1.80%). After pixel-level target centroid extraction and line-of-sight calculation for each frame of recognition results, the angle positioning error distribution under experimental conditions was statistically obtained. The results show that in the experiment... The mean of the orientation angle positioning error is -1.8 arcsec, the standard deviation is 9.10 arcsec, and the RMS is 9.3 arcsec. The mean of the orientation angle positioning error is 16.0 arcsec, the standard deviation is 18.3 arcsec, and the RMS is 24.0 arcsec. Comparison with simulation results shows that the error amplitude under experimental conditions is highly consistent with the expected value, and is generally smaller than the single-pixel field of view, corresponding to a ground projection error controlled within the range of approximately 20–60m. This result fully verifies the robustness and engineering feasibility of the algorithm of this invention in complex backgrounds and actual noise environments.

[0129] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for superimposed image target recognition and positioning based on sub-field shape modulation coding, characterized in that, Includes the following steps: Step 1: Foreground target identification; Identify foreground targets; Step 2: Target rotation state tracking; By fusing principal axis angle observation, angular velocity observation and second-order Kalman filtering, continuous estimation of target rotation is achieved; Step 3: Determine the correlation between the target's motion trajectory and the trend of light spot shape change, and inversely solve the actual spatial location; Key feature information of the target is extracted from each frame of image to further determine the target's motion trajectory and light spot morphology changes. Through dual-domain correlation modeling of target motion trajectory and light spot morphology evolution, the unique determination and high-precision positioning of the target's object space field of view are achieved. Step 3, the dual-domain correlation modeling of target motion trajectory and light spot morphology evolution, specifically includes: 3a) Modeling the motion of the center of mass; Let the target centroid in the frame is its inter-frame displacement is : ； in, For the first Frame target centroid; Inter-frame displacement Directional components; Inter-frame displacement Directional components; These correspond to the horizontal and vertical movement trends, respectively. 3b) Modeling of light spot morphology features; The definition of the goal is in the first The morphological parameter vector of the frame is: ； wherein: represents the pixel area of the target in the foreground mask; represents the scale of the spot in direction; represents the principal axis rotation angle obtained by the target rotation state tracking algorithm; Inter-frame difference characterizes the evolution trend of light spots: ； in, The change in the morphological parameter vector; For the first The morphological parameter vector of the frame; The change in pixel area of ​​the target in the foreground mask; For the light spot in The magnitude of change in direction; The change in the spindle rotation angle; Indicates clockwise rotation; Indicates counterclockwise rotation; 3c) Joint modeling of target motion trajectory and light spot morphology; In order to associate the target motion trajectory with the spot shape evolution, a joint feature vector is defined : ； The objective is to represent the overall evolution of the target within the timing window is represented as: ； in, For the target in the timing window The overall evolution within; For a certain moment The evolution of the target; Timing window At a certain moment within; 3d) Field of view region discrimination and inverse spatial location solution; The field of view is divided into several disjoint regions There is a deterministic mapping relationship : ； By identifying The target motion trajectory-spot morphology coupling feature contained in the middle directly determines the unique visual field area to which the target belongs ; According to the geometric correspondence between the field-of-view partition and the object space : ； Back-solving of targets from observations in the image plane to actual space coordinates .

2. The superimposed image object identification and localization method based on split field shape modulation coding according to claim 1, wherein, In step 3, the key feature information of the target is extracted, including: the position of the light spot centroid, the area of the pixels occupied, and the size of the target in the direction direction.

3. The superimposed image object identification and localization method based on split field shape modulation coding according to claim 1, wherein, Step 2 specifically includes: 2a) Observation of principal axis angle; For the target region detected in the foreground mask, the set of pixel coordinates within the region is recorded and centered; a covariance matrix is ​​constructed. By eigen decomposition , the largest eigenvalue of the covariance matrix corresponding eigenvector Target principal axis direction; principal axis angle is expressed as: ； wherein is a certain eigenvalue of the covariance matrix; is the eigenvalue corresponds to the eigenvector; is eigenvector in the direction; is eigenvector in the direction; 2b) Angular velocity observation; Select several stable feature points within the target area. Tracking is performed; feature points are calculated relative to the target centroid in each frame. Angle : ； in, Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target area. Moment coordinate; For the target center of mass Moment coordinate; Select several stable feature points within the target region coordinate; Select several stable feature points within the target region coordinate; and Representing the target centroid respectively coordinates and coordinate; This gives us the instantaneous angular velocity of the target. : ， ； in, For frame interval; For the first -1 frame feature points relative to the target centroid Angle; For the first Frame feature points relative to the target centroid Angle; Indicates the first One feature point, Indicates the total number of feature points; 2c) Second-order Kalman filter fusion; Will Provided principal axis angle observation and Using the provided angular velocity observations as input, a second-order Kalman filter is constructed to achieve joint estimation of the target rotation.

4. The superimposed image object identification and localization method based on split field shape modulation coding of claim 1, wherein, Step 1 specifically involves: using a Gaussian mixture model for background modeling, where the historical value of each pixel is modeled as a combination of multiple weighted Gaussian distributions, with each distribution corresponding to the background state of the pixel at different times; if the current pixel value deviates significantly from these background distributions, it is determined to be a foreground target.

5. The method of claim 1, wherein, Step 1 specifically involves using frame difference or optical flow methods to identify foreground targets.

6. The method for target recognition and localization of aliased images based on field-of-view shape modulation coding according to claim 1, characterized in that, In step 3: use The module extracts key feature information of the target in each frame of image.

Images